A Review of Deep Learning CT Reconstruction: Concepts, Limitations, and Promise in Clinical Practice

Abstract

Purpose of Review Deep Learning reconstruction (DLR) is the current state-of-the-art method for CT image formation. Comparisons to existing filter back-projection, iterative, and model-based reconstructions are now available in the literature. This review summarizes the prior reconstruction methods, introduces DLR, and then reviews recent findings from DLR from a physics and clinical perspective. Recent Findings DLR has been shown to allow for noise magnitude reductions relative to filtered back-projection without suffering from “plastic” or “blotchy” noise texture that was found objectionable with most iterative and model-based solutions. Clinically, early reader studies have reported increases in subjective quality scores and studies have successfully implemented DLR-enabled dose reductions. Summary The future of CT image reconstruction is bright; deep learning methods have only started to tackle problems in this space via addressing noise reduction. Artifact mitigation and spectral applications likely be future candidates for DLR applications.

Automatic Differentiation for Inverse Problems in X-ray Imaging and Microscopy

“…X-ray Computer tomography (CT) [ 59 , 60 ] is used to analyse biological samples in their native environment [ 61 ] in a non-destructive way. In its simplest form [ 62 ], we aim at reconstructing a 3D volume starting from a series of 2D projections , acquired at different rotation angles , where i denotes the i th observation [ 1 ].…”

“…(i) Radiation damage sets limits on the beam intensity for each projection [ 68 ]; when the number of photons is scarce (e.g., due to dose fractionation criteria [ 69 ] or photon starvation [ 70 ]), common single-shot algorithms produce very noisy images [ 60 ]. Ex-post noise removal increases the output Signal-to-Noise-Ratio (SNR) at the expense of spatial resolution.…”

“…The problems discussed so far produce very noticeable and characteristic artefacts in the reconstructions: while FBP output images are noisy, have poor contrast and are blurry [ 60 ], iterative methods might generate sub-optimal image texture, often referred to as “plastic” or “blotchy” [ 78 ]. In this context, AI can provide a huge help, as it can learn the correspondence between the latent object and the characteristic artefact/defect triggered in a well-defined reconstruction algorithm by a well-defined problem [ 79 ].…”

“…In this context, AI can provide a huge help, as it can learn the correspondence between the latent object and the characteristic artefact/defect triggered in a well-defined reconstruction algorithm by a well-defined problem [ 79 ]. A review of methods is in [ 60 ].…”

Improved artificial intelligence discrimination of minor histological populations by supplementing with color-adjusted images

“…Since the last decade of the 20 th century, innovative advances in computer and digitization technologies have initiated methods for the digital storage and utilization of a variety of medical information, especially medical images 1 . Further advances in medical engineering have enabled the usage of medical instruments such as endoscopy 2 , computed tomography 3 , and magnetic resonance imaging 4 in medical practice. Recent developments have focused on modern computer-aided diagnoses relying on advancements in artificial intelligence (AI) technology.…”